AVIA — Audio-Visual Integrity Analyzer for Multimodal Deepfake Detection
Designed and developed a multimodal deepfake-detection framework integrating ConvNeXt for image forensics, GenConViT for video analysis, and a CNN–BiLSTM architecture for synthetic-speech detection. The system provides one verification workflow across images, videos, and audio.
Trained and evaluated the framework on 12 benchmark datasets containing more than 5,000 samples. Built specialised pipelines to distinguish authentic content from GAN-generated and diffusion-generated media and to evaluate robustness across modalities.
Thesis document ↗
Smart Attendance System with Multi-Face Recognition
Developed a Python backend service for real-time multi-face recognition using MTCNN and SVM models, achieving 91% detection accuracy and processing times below 850 milliseconds per image.
Built a cross-platform application around a Flask architecture, enabling consistent real-time image analysis across devices and supporting attendance capture for multiple faces within a single frame.
Read on Medium ↗
CI/CD Pipeline for Containerised Application Deployment
Reduced manual deployment time by 80% by orchestrating a Jenkins-based continuous-delivery pipeline for containerised React and Node.js services across development, testing, and staging environments on AWS EC2.
Configured separate frontend, backend, and database containers, established shared security groups and IAM roles, and implemented automated unit and integration pipelines to maintain environment parity and reliable container-to-container communication.
Hybrid RFID–PIN Door-Lock System
Built a double-layered access-control system combining RFID verification with PIN authentication. The project demonstrates how two independent authentication factors can strengthen physical security while remaining practical for low-cost deployment.
View publication ↗
